Warnings Raised Over AI Training Data and Chatbot Trust

AI 안전 | Sun Jun 21 2026 00:00:00 GMT+0000 (Coordinated Universal Time) | 2 sources

The Atlantic exposed unauthorized use of music datasets for AI training while Signal's CEO warned about chatbot privacy risks.

Analysis

[The Atlantic] published a searchable database of AI music training datasets ^[1]

• Journalist Alex Reisner uncovered four datasets
• Largest datasets contain up to 12 million and 9 million songs
• Google and Stability confirmed usage in research papers
• Includes Lady Gaga
• Radiohead
• Wu-Tang Clan and others

[AI training data collection practices] exposed use of tools that bypass platform terms of service ^[1]

• Three datasets distributed as YouTube/Spotify links
• Tools bypass login
• ads
• and monetization mechanisms
• Free Music Archive requires license for commercial use
• Downloaded thousands of times suggesting widespread use

[Signal (Meredith Whittaker)] warned that AI chatbots are not friends ^[2]

• Emphasized ChatGPT and Claude are not conscious entities
• Warned thinking process risks being obscured by system responses
• Uses AI only for tasks like document formatting
• Expressed privacy concerns in Bloomberg interview

[Microsoft Copilot] raised backdoor concerns over AI agents' broad access permissions ^[2]

• Criticized Mustafa Suleyman's Christmas shopping automation scenario
• Requires access to credit cards
• browser
• Signal
• and calendar
• Scenario involves eavesdropping on family group chats for gift decisions
• Could effectively act as a backdoor in Signal's context

Sources

• [1] The Atlantic created a searchable database of the music used to train AI - The Verge AI
• [2] Signal’s Meredith Whittaker wants you to remember that AI chatbots ‘are not your friends’ - TechCrunch AI